Research map / Label-efficient and robust learning / Self-supervised and contrastive
How robust is unsupervised representation learning to distribution shift?
Yuge Shi, Imant Daunhawer, Julia E Vogt, Philip Torr, Amartya Sanyal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use SSL (avg of SimCLR, SimSiam, BYOL) instead of supervised learning | representation learner · MNIST-CIFAR CIFAR label | OOD accuracy: 51.5 -> 86.1 | Improved |
| use autoencoder family (AE, VAE, IWAE, beta-VAE) instead of supervised learning | representation learner · MNIST-CIFAR CIFAR label | OOD accuracy: 51.5 -> 79.9 | Improved |
| use SSL instead of supervised learning under extreme shift r_id=1 | representation learner · CdSprites shape | OOD accuracy: 44.0 -> 82.7 | Improved |
| use AE instead of supervised learning under extreme shift r_id=1 | representation learner · CdSprites shape | OOD accuracy: 44.0 -> 52.5 | Improved |
| use SSL instead of supervised learning | representation learner · Camelyon17 | OOD accuracy: 86.8 -> 89.8 | Improved |
| use SSL instead of supervised learning | representation learner · FMoW worst-group | OOD accuracy: 35.6 -> 29.6 | Hurt |
| use AE instead of supervised learning | representation learner · FMoW worst-group | OOD accuracy: 35.6 -> 22.8 | Hurt |
| use SSL instead of supervised learning | representation learner · FMoW worst-group | shift sensitivity: 37.7 -> 9.1 | Improved |
| use SSL instead of supervised learning | representation learner · MNIST-CIFAR CIFAR label | shift sensitivity: 47.6 -> 13.7 | Improved |
| retrain linear head on OOD data instead of ID data (ablation) | SSL representations · MNIST-CIFAR CIFAR label | OOD accuracy (linear head bias b): 34.8 | Improved |
| retrain linear head on OOD data instead of ID data (ablation) | supervised representations · MNIST-CIFAR CIFAR label | OOD accuracy (linear head bias b): 0.8 | No clear effect |
| retrain linear head on small OOD data (10% of OOD test) (ablation) | AE and SSL · Camelyon17 | OOD accuracy (linear head bias b): 31.7 (AE), 8.8 (SSL) | Improved |
Closest papers by meaning
- Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations (NeurIPS 2024 · similarity 0.88)
- Tolerant Algorithms for Learning with Arbitrary Covariate Shift (NeurIPS 2024 · similarity 0.86)
- Learning Neural Networks with Distribution Shift: Efficiently Certifiable Guarantees (ICLR 2025 · similarity 0.86)
- Towards Robustness to Natural Variations and Distribution Shift (Student Abstract) (AAAI 2024 · similarity 0.85)
- Transformation-Invariant Learning and Theoretical Guarantees for OOD Generalization (NeurIPS 2024 · similarity 0.85)
- Sufficient Invariant Learning for Distribution Shift (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
